Spatiotemporal Analysis of Land Surface Plant Phenology

Abstract Plant phenology is the study of cyclic and seasonal plant life developmental stages and is used by scientists and policy-makers to understand agricultural management practices, climate change effects, and ecosystem processes. In particular, land surface phenology relates to the seasonal pattern of variation in vegetated land surfaces where measurements are typically collected via remote sensing. In the past, estimates of plant phenometrics required a substantial amount of data, sometimes aggregated over many years or pixels of satellite imagery. Recent work has drawn on the power of Bayesian hierarchical modeling (BHM) applied at the pixel level to reliably quantify complete annual time series of phenometrics despite sparseness in the temporal domain. This work extends these approaches by jointly modeling phenology across pixels to account for spatial dependencies and introduces a layer to the BHM to quantify trends in phenology. We assume the phenometrics follow a nonlinear double-logistic function of time and propose a linearized approximation to facilitate computation for a large spatial domain. We use a simulation study to verify the conditions under which this approximation gives valid statistical inference and then use the method to estimate phenometrics and identify temporal trends in the Harvard Experimental Forest with data collected by the United States Geological Service and the National Aeronautics and Space Association (USGS/NASA) Landsat satellite program from 1984 to 2020.